I just don't understand how come this is so much of an oversight, neural networks today can synthesise voice, pictures. ALSO neural networks today can reliably

I just don't understand how come this is so much of an oversight, neural networks today can synthesise voice, pictures. ALSO neural networks today can reliably detect if a picture or voice was generated by neural networks. How come someone skipping the latter in a security application? A BANK? I just don't understand.